hsaest/Llama-3.1-8B-Instruct-travelplanner-SFT

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:32kPublished:Oct 23, 2024Architecture:Transformer0.0K Warm

The hsaest/Llama-3.1-8B-Instruct-travelplanner-SFT is an 8 billion parameter instruction-tuned model based on the Llama 3.1 architecture, specifically fine-tuned for planning tasks. With a 32768 token context length, this model demonstrates improved performance in commonsense and hard planning scenarios compared to its base model and Qwen2-7B, particularly after fine-tuning. It is optimized for complex reasoning and sequential decision-making, making it suitable for applications requiring robust planning capabilities.

Loading preview...

Model Overview

The hsaest/Llama-3.1-8B-Instruct-travelplanner-SFT is an 8 billion parameter instruction-tuned model built upon the Llama 3.1 architecture. It features a substantial context length of 32768 tokens, enabling it to process and understand extensive planning-related inputs. This model has been specifically fine-tuned to enhance its performance in various planning tasks, as detailed in the provided benchmark results.

Key Capabilities

  • Enhanced Planning Performance: The model shows significant improvements in both commonsense and hard planning tasks after fine-tuning, outperforming its base Llama 3.1-8B and Qwen2-7B counterparts.
  • Robust Commonsense Reasoning: Achieves a 78.3% score in Commonsense (Micro) and 17.8% in Commonsense (Macro) after fine-tuning, indicating strong understanding of everyday planning logic.
  • Improved Hard Planning: Demonstrates a 19.3% score in Hard (Micro) and 6.1% in Hard (Macro) after fine-tuning, suggesting better handling of more complex and challenging planning problems.
  • Higher Final Pass Rate: Achieves a 3.8% final pass rate after fine-tuning, a notable improvement over direct prompting and other models in its class.

Good For

  • Travel Planning Applications: Its specialization in planning makes it highly suitable for generating itineraries, suggesting routes, and managing travel logistics.
  • Complex Reasoning Tasks: Ideal for scenarios requiring sequential decision-making and logical progression.
  • Agent-based Systems: Can serve as a core component for AI agents that need to plan actions and strategies.
  • Research in Planning: Useful for researchers exploring the capabilities and limitations of language models in planning domains, as highlighted by the associated research paper: Revealing the Barriers of Language Agents in Planning.